vtk-cmake

CMake configurations for how I like to build VTK.

Pull Pre-built Images

I've built Docker images for a few Python and VTK versions and pushed them to the following images in this repositories container registry:

• ghcr.io/banesullivan/pyvista: PyVista configured for JupyterLab (VTK OSMesa)
• ghcr.io/banesullivan/pyvista-gpu: PyVista configured for JupyterLab (VTK EGL for use with NVIDIA GPUs)
• ghcr.io/banesullivan/python-vtk: Python slim image with VTK (OSMesa) installed.

The tags for these images follow the format {Python Version}-{VTK version}. For example, the image with Python version 3.11 and VTK version 9.3.0 would be: ghcr.io/banesullivan/pyvista:3.11-9.3.0

You can pull this image from the container registry here:

docker pull ghcr.io/banesullivan/pyvista:3.12-9.3.1

To run it:

docker run -it -p 8888:8888 --rm ghcr.io/banesullivan/pyvista:3.12-9.3.1

There is also a GPU version ready for use with NVIDIA hardware:

docker pull ghcr.io/banesullivan/pyvista-gpu:3.12-9.3.1
docker run -it -p 8888:8888 --rm --runtime=nvidia --gpus all ghcr.io/banesullivan/pyvista-gpu:3.12-9.3.1

Build it yourself

I have set up a Docker build to generate a VTK wheel and install that wheel along with PyVista in a Docker image with JupyterLab ready to go. There are two targets for the build:

• slim: Builds and installs VTK into a python-slim image (without PyVista)
• jupyter: Builds and installs VTK with Python and JupyterLab

To build the slim image

docker build --target slim -t ghcr.io/banesullivan/python-vtk .

Or you can specify specific versions of Python and VTK by:

# Build with Python 3.12
docker build --target slim --build-arg='PYTHON_VERSION=3.12' -t ghcr.io/banesullivan/python-vtk .

# Build VTK 9.3 with EGL (defaults to Python 3.12)
docker build --target slim --build-arg='VTK_VERSION=v9.3.1' --build-arg='VTK_VARIANT=egl' -t ghcr.io/banesullivan/python-vtk .

After building, you can extract the wheel or use that base image for all your VTK Python needs.

The wheel is saved uner /opt/vtk/

To run a JupyterLab instance with EGL and access to NVIDIA GPUs:

# Build VTK 9.3 with EGL for use in Jupyter
docker build --target jupyter --build-arg='VTK_VERSION=v9.3.1' --build-arg='VTK_VARIANT=egl' -t ghcr.io/banesullivan/pyvista-gpu .

# Run the image with NVIDIA runtime and GPU access
docker run -it -p 8888:8888 --rm --runtime=nvidia --gpus all ghcr.io/banesullivan/pyvista-gpu

PyVista/VTK with OSMesa (CPU-rendering) in JupyterLab:

docker build --target jupyter --build-arg='VTK_VERSION=v9.3.1' --build-arg='VTK_VARIANT=osmesa' -t ghcr.io/banesullivan/pyvista .

docker run -it -p 8888:8888 --rm --runtime=nvidia --gpus all ghcr.io/banesullivan/pyvista

PyVista/VTK with EGL (GPU-rendering) in JupyterLab:

docker build --target jupyter --build-arg='VTK_VERSION=v9.3.0' --build-arg='VTK_VARIANT=egl' -t ghcr.io/banesullivan/pyvista-gpu .

docker run -it -p 8888:8888 --rm --runtime=nvidia --gpus all ghcr.io/banesullivan/pyvista-gpu

Manual Build

1. Copy/paste the *.cmake files in the cmake/ directory here to your build directory.
2. Choose a build configuration. Each configuration is a .cmake file without the _ prefix.
   • For example, to build the OSMESA variant, choose the osmesa.cmake configuration
3. Configure the build using the chosen configuration:

   cmake -GNinja -C osmesa.cmake ..
   

4. Build!

   ninja
   

5. Make a wheel (will then be generated in VTK/build/dist)

   python setup.py bdist_wheel
   

For example:

git clone https://github.com/Kitware/VTK.git
cd VTK
git checkout release
mkdir build
git clone https://github.com/banesullivan/vtk-cmake.git
cp vtk-cmake/cmake/*.cmake build/
cd build
cmake -GNinja -C osmesa.cmake ..
ninja
python setup.py bdist_wheel

Reference scripts/build.sh

Contributing

Feel free to add your own configurations that adopt from the base configurations _base.cmake and _modules.cmake.

Modules

To turn on/off modules not included here, specify these in the _modules.cmake configuration file.

The enabled/disabled modules follow a similar specification to the default for what VTK publishes on PyPI. Reference:

• https://gitlab.kitware.com/vtk/vtk/-/blob/master/.gitlab/ci/configure_macos.cmake

This has the behaviour of "enable all, disable some".

Additionally Qt and MPI are disabled in these configurations.

Some additions

• Module_vtkAcceleratorsVTKm:BOOL=OFF

Further Resources

See VTK's build notes

Linux Build Dependencies

I think these are all the system dependencies needed:

sudo apt-get update
sudo apt-get install --no-install-recommends --yes \
  build-essential \
  libosmesa6-dev \
  cmake \
  ninja-build \
  libboost-all-dev

And Python of course :)

Windows Notes

Some reminders for how to build on windows:

• Use "x64 Native Tools Command Prompt" for the installed Visual Studio version to configure with CMake and to build with ninja.
• Many of the necessary Python paths are hardcoded in the configurations (for my local machine) and need to be updated per-environment.

OpenVR

• VTK_BUILD_PYI_FILES must be set to OFF
• Copy openvr_api.dll to the installed vtkmodules path